include end-to-end convolution tests
include scaling and componentwise product routines for convolution
write man pages
write fftc8_pass in asm to schedule it properly; or find a better compiler
take advantage of sqrt(1/2) (1+i) halfway through pass loop
write fftc8_8 from scratch
maybe write fftc8_16 from scratch
speed up fftc8_512, fftc8_1024
include larger transforms?
eliminate overlap between fftc8_2, fftc8_un2, small root table
